Uma plataforma para monitoramento de espectros sonoros e pré-processamento de canto de pássaros

Abstract

The current methods for birdsong automatic classification present gaps which justify a deeper study of the theme. Among these shortcomings, for example, there is the need for the development of research for a wider range of species. Another gap that will be the focus of this work, is observed in the initial stages of the process of automatic classification in real environments and practical (e. g. nature). Considering that the stages of sample collection, preprocessing, processing and analysis of results are present in a birdsong classification process, there is a deficiency of contributions in the stage of sample collection and, particularly, in the preprocessing step. In this context, the routines designed on the scientific papers of sound classification are often dependent on an ideal sample, which are not found in a real environment. That is, to achieve correct answers and successes in classification such routines require filtered samples and also a scope of environment previously restricted. With the objective of providing solutions to this set of deficiencies, it was proposed and implemented a platform for collecting and processing of sounds collected, covering the first two steps of automatic classification. In summary, this paper presents a platform for acquisition and preprocessing of bird songs. On paper are also discussed about the advantages and disadvantages of the methodologies used, as well as the contexts relevant to the application of the platform. The validation occurs through practical tests and comparisons between the samples processed by the platform, with samples arising from the previously existing tools of sound processing and analysis.